Speech Disfluencies in Spoken Language Systems: A Dialog- Centered Approach

这是一个为期三年的持续研究项目的第一年资助，该项目旨在研究如何利用口语对话系统中的不流利现象，以及如何提高计算机化对话系统处理不流利现象的能力。在第一个任务中，研究了应该编辑或保留流利对话的哪些部分，以及它对预测说话者计划问题的影响。在第二项任务中，调查的中心是系统利用对话中的多模态反馈来修复由不流畅引起的误解的能力。此外，对语料库进行了分析，目的是为口语系统建立一个包含系统从故障中恢复能力的对话评估指标。

This is the first year funding of a three-year continuing award to investigate how to exploit disfluencies in spoken language dialog systems and also how to enhance the ability of computerized dialog systems to handle disfluencies. In the first task the decision of what parts of the diffluent dialog should be edited or retained is studied as well as its influence in anticipating the speakers' planning problems. In the second task the investigation centers on the systems' ability to utilize multimodal feedback adapted from conversation to repair misinterpretations arising from disfluencies. Additionally, corpora is analyzed with the objective of establishing a dialogic evaluation metric for spoken language systems which incorporates the system's ability to recover from faults.